GOAT
저희 교수님이십니다.
오병우
Lab Head
The importance of spatio-temporal trajectories for contact tracing has increased due to the recent COVID-19 pandemic. Spatio-temporal trajectories store time and spatial data of moving objects. In this paper, I propose query processing for spatio-temporal trajectories of moving objects. The spatiotemporal trajectory model of moving objects has point type spatial data for storing locations and timestamp type temporal data for time. A trajectory query is a query to search for pairs of users who have been in close contact by boarding the same bus. To process the trajectory query, I use the Geolife dataset provided by Microsoft. The proposed trajectory query processing method divides trajectory data by date and checks whether users’ trajectories were nearby for each date to generate information about contacts as the result.
Applications such as digital twins, urban planning, and facility management use spatial data consisting of geometric coordinates and additional non-spatial attributes. Traditional spatial indexes search for geometric coordinates, requiring a separate filtering process for non-spatial attributes. This paper proposes NR*-tree(Non-spatial and Spatial Region Star Tree), integrating non-spatial attributes into a spatial index to eliminate the filtering step and enhance search efficiency. Numerical non-spatial attributes extend the dimensions of MBRs(Minimum Bounding Rectangles) to store ranges, while categorical non-spatial attributes store single values in MBRs for filtering at leaf nodes. Experiments show that the NR*-tree reduces search time by up to 86.8% compared to the conventional R*-tree (Region Star Tree). Including the length of the major axis in MBRs during searches enhances map readability and further improves search performance.
This paper proposes a method for classifying spatial data, which is increasingly important as a base technology for mobile environments and digital twins, using deep learning technology. CNN is used as a deep learning technology for spatial data classification. A dataset for learning is constructed by preprocessing spatial data. The learning model for training and validating the dataset consists of a convolution layer and a fully connected layer. The convolution layer adopts the convolution layer of VGG16 and uses transfer learning. The fully connected layer consists of a Flatten layer, a ReLU function, a Dropout layer, and a Softmax function. An experiment in which the spatial data classification method proposed in this paper was applied to Korea national map was performed, and verification accuracy of 99.1% was obtained.
